Hydrogeological responses to the 2016 Gyeongju earthquakes , 1 Korea 2

Abstract. The September 12, 2016 Gyeongju earthquakes (M5.1 and M5.8) had significant effects on groundwater systems along the Yangsan Fault System composed of NNE-trending, right-lateral strike-slip faults in Korea. Hydrological changes induced by the earthquakes are important because no surface ruptures have been reported and few earthquakes usually occur in Korea. The main objective of this research was to propose a conceptual model interpreting the possible mechanisms of groundwater response to the earthquakes based on anomalous hydrogeochemical data including isotope (radon, strontium) concentrations with bedrock characteristics. To analyze the hydraulic changes resulting from the earthquakes, annual monitoring data of groundwater level, temperature, and electrical conductivity and collected data of hydrochemical parameters, radon-222, and strontium isotopes were collected during January 2017. Groundwater level anomalies could be attributed to the movement of the epicentral strike-slip fault. Radon concentration data showed the potential of groundwater mixing processes. Strontium anomalies could be related to the lithology and stratigraphy of the bedrock, reflecting the effect of water–rock interaction. Using a Self-Organizing Map (SOM) statistical analysis, associations of hydro-geochemical characteristics among groundwater wells were interpreted. By combining the grouped results of the SOM with lithostratigraphic unit data, 21 groundwater wells were classified into four groups, each corresponding to different hydrogeological behaviors. A new comprehensive conceptual model was developed to explain possible mechanisms for the hydrological and geochemical responses in each group, which have been respectively identified as water–rock interaction, mixing of shallow and deep aquifers via sea water intrusion, bedrock fracture opening related to strike-slip fault movement, and no response.



33
Earthquakes have a great influence on groundwater hydrology, such as water table changes 34 and hydrochemical anomalies. Typically, most studies have focused on earthquake 35 forecasting, i.e. changes prior to earthquakes, or co-seismic behavior. There have been a 36 limited number of studies that discuss the responses of groundwater systems, especially, 37 focused on the hydrogeologic changes after earthquakes (Adinolfi Falcone et al., 38 2012;Amoruso et al., 2011;Barberio et al., 2017;Claesson et al., 2007;Ekemen Keskin, 39 2010; Galassi et al., 2014;Lee et al., 2013;Matsumoto et al., 2003;Petitta et al., 2018;Wang 40 and Manga, 2010;Wang et al., 2012;Yechieli and Bein, 2002). Seismic waves, for example, 41 are known to cause changes in water level, temperature, and geochemistry (Matsumoto et al., 65-74º with a depth ranging from 11 km to 16 km, The width of the distribution of event 148 locations is approximately 5 km in length, and was determined to be a branch of the YSF 149 (Hong et al., 2017;Kim et al., 2017a;Lee et al., 2018;Son et al., 2017). 150 Twelve wells are located near the YSF and the surrounding area within the Gyeongsang 151 Basin (Fig. 2b). The information for each well is shown in Table 2. The wells generally were 152 installed by two types at each point. The one well (as labeled KW ##) indicates that the 153 sampling point is consisted of only one type well, bedrock aquifer well. The alluvial aquifer 154 wells were labeled KW##-1 and the bedrock aquifer wells were labeled KW## or KW##-2. 155 The KW 11-3 refers the surface water sample near the KW 11 well. The lithostratigraphic 156 unit indicates the characteristic of the bedrock aquifer wells (labeled as KW##-2). The consist of Early and Middle Miocene sedimentary and volcanic rocks that are 'mainly 168 exposed in the eastern part of the Gyeongsang Basin. The Yeonil group basin consists of a 169 tuffaceous Tertiary sedimentary basin, and Miocene basal conglomeratic rocks, which consist 170 of light brown to light gray conglomerate and sandstone. The Janggi group rocks mainly 171 consist of basaltic tuff and andesitic tuff of Early Miocene age. The Bulguksa intrusive rocks 172 are mainly composed of biotite granites accompanying grano-diorite, tonalite, and alkali-173 feldspar granites (Hwang et al., 2004). Based on the lithology and stratigraphy, this study 174 divided the bedrock aquifer well locations into four areas; (i) Hayang-group shale and 175 sandstone (KW 1, KW 2, KW 9-2, KW 10-2), (ii) Bulguksa-group biotite granite (KW 3, KW 176 5-2, KW 12-2), (iii) tuff and tuffaceous sedimentary rocks of Yeonil-group and Janggi group 177 (KW 4-2, KW 6-2, KW 7-2), and (iv) Cretaceous volcanic rocks mainly composed of earthquake. Precipitation data obtained from the Korea Meteorological Administration were 187 also analyzed with the water level variation (http://kma.go.kr). These daily data correspond to 188 the cumulative quantities during the day.

189
Sampling of groundwater wells in alluvial and bedrock aquifers was conducted for three 190 days (January 16, 2017 to January 18, 2017) four months after the earthquake. A total of 22 191 water samples, including one of surface water, were collected in 2-L polyethylene bottles 192 using a Grundfos MP-1 pump. The samples were analyzed for hydrochemical parameters, 193 major ions, radon concentration, and strontium isotopes. The hydrochemical parameters 194 temperature, EC, dissolved oxygen (DO), total dissolved solids (TDS), pH, and salinity were 195 measured in the field using an YSI ProDSS digital sampling system (Xylem, USA). The 196 analysis of cations and anions (Na,K,Ca,Mg,Cl,NO 3 ,SO 4 , and HCO 3 ) including strontium 197 isotopes, was completed using filtered water samples at the Korea Basic Science Institute recovered to the original value after an instantaneous increase, whereas that of KW 11-2 245 recovered after a slight decrease. This anomaly was apparent in the alluvial aquifer well (KW 246 11-1), unlike the groundwater level anomaly.

247
A change in groundwater ECs was observed at eight monitoring wells before, during, or 248 after the earthquake. KW 1 responded to the earthquake in a peak form and gradually 249 recovered. KW 2 showed an increase prior to the earthquake and recovered to original values.

250
The groundwater EC consistently decreased and then remained at lower values at KW 6-1, 251 KW 6-2, and KW 10-2. The peak form of the KW 11 wells also indicated an opposite 252 direction (Fig. 4b). KW 11-1 peaked at a higher level several times prior to the earthquake, 253 while KW 11-2 peaked at a lower level before the earthquake and then recovered. Compared 254 to the water level data, however, it was difficult to interpret that the changes in EC could be 255 attributed to the earthquake.

257
The hydrogeochemical data of 17 parameters (Na,K,Ca,Mg,Cl,NO 3 ,SO 4 , HCO 3 , 258 temperature, pH, DO, EC, TDS, salinity, Sr, 87 Sr/ 86 Sr, and radon) were collected from 21 259 groundwater samples and one surface water sample (KW 11-3) in January 2017. The 260 analytical results of the water samples are summarized in Table 3. The Na values were high in 261 The SO 4 values were high in KW 1 and KW 10-2 (> 200 mg/L).

264
The distribution of radon concentration in the 21 groundwater samples is shown in Fig. 5.

285
The spatial distributions of strontium concentrations and 87 Sr/ 86 Sr are shown in Fig. 7.

Self-Organizing Map (SOM)
There are few studies using SOM for groundwater quality data interpretation (Choi et al., 302 2014;Hong and Rosen, 2001;Lischeid, 2008). However, we used the SOM analysis for 303 statistical analysis in the Gyeongju area because it can solve linear dimensionality reduction 304 problems without biases. This method also provides the detailed local relationship between 305 the variables by the component planes, which is helpful to understand groundwater systems  Table 3). The raw data were normalized in 311 order to work with transformed quantities with zero mean and unit standard deviation. By

318
The clustering could be investigated with the visual inspection of the U-matrix result ( Fig.   319 10). Deep brown shades on the U-matrix indicate a large distance between neighborhood 320 nodes whereas white shades correspond to a short distance between nodes. Based on the distances, the distribution of water samples could be classified into four groups: Group 1 322 (KW 1, KW 2, KW 9-1, and KW 10-1), Group 2 (KW 3, KW 5-1, KW 5-2, KW 6-2, KW 11-323 3, and KW 12-1), Group 3 (KW 4-1 and KW 4-2), and Group 4 (KW 8-1, KW 11-1, and KW   Table 4. The groundwater level, 335 temperature, and EC data were analyzed considering pre-, co-, and post-seismic changes. For 336 the groundwater level data, three anomaly types were observed (see Fig. 3). Among them, the 337 maintenance of a groundwater level increase could be attributed to aquifer compaction (as 338 observed in KW 8-1 and KW 8-2) (Lee et al., 2002). There is a possibility that the aquifers 339 underwent non-recoverable deformation. The persistent groundwater level changes also have been influenced by volumetric strain changes (Matsumoto et al., 2003;Roeloffs et al., 341 2003;Wang et al., 2007). In contrast, a greater decrease in groundwater level prior to the 342 earthquake could be attributed to the opening of bedrock fractures (as observed in KW 11-1) 343 (Fleeger et al., 1999;Kitagawa et al., 2006;Rojstaczer and Wolf, 1992;Rojstaczer et al., 344 1995;Wang et al., 2004). A decrease could also possibly be related to a change in 345 permeability (Brodsky, 2003;Manga and Wang, 2007). Groundwater level oscillation also 346 depends on the interactions between inflow/outflow of the well and of the aquifer (Cooper et 347 al., 1965). There is another anomaly, an opposite change pattern between the alluvial and 348 bedrock aquifer wells, as observed in all datasets of groundwater level, temperature, and EC 349 data for KW 11 (see Fig. 3 and Fig. 4). This means that the two wells had weak interactions 350 with each other.

352
The hydrogeochemical data of 17 parameters (Na, K, Ca, Mg, Cl, NO 3 , SO 4 , HCO 3 , 353 temperature, pH, DO, EC, TDS, salinity, Sr, 87 Sr/ 86 Sr, and radon) were collected only after 354 earthquake (Jan., 2017). A difference in radon concentration between the alluvial and bedrock 355 aquifer wells could be considered more significant because of the mixing effect as observed 356 in KW 8 and KW 11 (see Fig. 5). Seismotectonic activity may often change the mixing ratio 357 of groundwater in a well (Claesson et al., 2007;Hartmann and Levy, 2005). The anomaly in 358 which the alluvial aquifer well had a higher radon concentration than that of the bedrock aquifer could be attributed to rainfall; however, in this area, rainfall did not occur during the

379
This paper is to combine the hydrologic, hydrogeochemical, and lithostratigraphic 380 characteristics by applying the neural network method SOM. The SOM results using the 381 hydrogeochemical data showed the 4 groups, however, these did not include KW 6-1, KW 7-  Fig. 9). These correlations were also used for analyzing the possible mechanisms at each 397 group. The piper diagram was also analyzed with the groups (Fig. S1). In this diagram, most water samples were prevailing SO 4 and HCO 3 , suggesting the possibility of water body 399 mixing with other water type.

400
The lithology and stratigraphy of Group A is classified as Hayang group shale and sandstone 401 of low porosity and high strontium concentrations. Particularly, the KW 9 and the KW 10 402 wells had a low radon concentration (< 1000 Bq/m3), high strontium concentration, high Ca 403 value, low 87 Sr/ 86 Sr ratio and low pH (see Fig. 8 and Fig. 9). There might be some possible the SOM results, KW 1, KW 2, KW 9-1, and KW 10-1 were clustered as one group (see Fig.   420 10), also suggesting the strong influence from bedrock to shallow aquifer.

421
Group B wells are located in biotite granitic region of the Bulguksa group, which has a 422 typical high radon concentration. The radon concentration is greatly influenced by uranium 423 content; thus, its concentration is generally high in granite compared to that of sedimentary 424 rocks. Typically, uranium concentration is high in granites, whereas it is low in sedimentary 425 rocks. However, only the KW 5 wells had a high radon concentration. In particular, KW 5-1 426 had high values similar to those of KW 5-2. This could be attributed to deep fluid upwelling 427 from the bedrock in the KW 5 wells (Chiodini et al., 2000;Minissale, 2004;Savoy et al., 2011).

430
Group C is composed of tuff and tuffaceous sedimentary rocks of the Yeonil and Janggi 431 groups. This group had a low radon concentration and a small difference in radon 432 concentration between the alluvial and the bedrock aquifer wells (see Fig. 5), suggesting 433 active water mixing between the two aquifers. In addition, the bedrock of this area contains 434 conglomerates, which generally have high pore density, leading to active mixing with water 435 compared to the shale-dominant lithology. This hypothesis seems to be consistent with the 436 weak chemical signature of Group A. KW 4-1, KW 4-2, and KW 6-2 had high values of EC, Cl, TDS, and salinity values (see Table 3 and Fig. 9), suggesting the possibility of sea water 438 intrusion by the effects of the earthquakes. The piper diagram also indicated the sea water 439 intrusion by the location of Group C, which got toward Na + +K + and HCO 3 -(see Fig. S1).

440
These water samples were prevailing HCO 3 water. Sea water intrusion might actively trigger 441 mixing between the shallow and deep aquifers.

442
In Group D, the radon concentration was quite different between the two wells and the 443 groundwater level anomaly occurred (see Fig. 3  Ca 2+ and Sr 2+ with a low 87 Sr/ 86 Sr ratio (see Fig. 8 and Fig. 9) (Bullen et al., 1997;Franklyn et 458 al., 1991;Négrel, 2006). In contrast, one groundwater chemistry study in Canada showed that 459 dissolution of alkali feldspar can increase the 87 Sr/ 86 Sr ratio providing sodium and potassium 460 (Bullen et al., 1996). Therefore, the various compositions of the granite and the fluid mobility 461 would be determinative in the 87 Sr/ 86 Sr ratio. Moreover, KW 11-1 also located in high SO 4 2-462 +Clthan KW 11-2 in the piper diagram, suggesting the deep water influence to KW 11-1 463 (Reddy and Nagabhushanam, 2012).

464
In accordance with this analysis, conceptual models of groundwater changes induced by the 465 earthquakes can be suggested (Fig. 11). Four different models were inferred by data analysis